gistfile1.irclog

10:39 < major> well, it goes something like this: ssh-keygen -f 
               servers_CA;ssh-keygen -f users_CA
10:40 < major> after that: ssh-keygen -s <CA_file> -I <description> -n 
               <principals> <pubkey_file>
10:40 < major> for users
10:40 < major> and: ssh-keygen -s <CA_file> -I <description> -h -n <hostnamess> 
               <pubkey file>
10:40 < major> for signing host pubkeys
10:41 < major> you can add a @ca-authority in your authorized keys to trust all 
               signed keys by that CA
10:41 < major> erm
10:41 < major> to known_hosts
10:42 < major> like, take the users_CA.pub and add it to known_hosts: 
               @cert-authority *.example.com ssh-rsa .....
10:42 < major> suddenly no more "do you want to trust this server" when you ssh 
               to a host with a signed cert
10:43 < major> the /etc/ssh/sshd_config is a little more verbose for the setup
10:43 < major> you need a HostCertificate entry per HostKey
10:43 < major> HostKey /etc/ssh/ssh_host_rsa_key
10:43 < major> HostCertificate /etc/ssh/ssh_host_rsa_key-cert
10:43 < major> HostKey /etc/ssh/ssh_host_dsa_key
10:43 < major> HostCertificate /etc/ssh/ssh_host_dsa_key-cert
10:43 < major> HostKey /etc/ssh/ssh_host_ecdsa_key
10:43 < major> HostCertificate /etc/ssh/ssh_host_ecdsa_key-cert
10:43 < major> HostKey /etc/ssh/ssh_host_ed25519_key
10:43 < major> HostCertificate /etc/ssh/ssh_host_ed25519_key-cert
10:43 < major> when you sign a servers ssh_host_rsa_key.pub you will get an 
               ssh_host_rsa_key-cert.pub
10:44 < major> you just copy it back to the server and fixup the entries
10:44 < major> the last 2 lines you the sshd_config are the kicker
10:44 < major> TrustedUserCAKeys /etc/ssh/users_CA.pub (from before)
10:44 < major> and: AuthenticationMethods publickey,password 
               publickey,keyboard-interactive
10:45 < major> setting AuthenticationMethods lets you decide what is required 
               in an auth set, and the order they are required in
10:45 < major> spaces separate sets, commas separate methods that have to 
               succeed in that set
10:45 < major> the entire set has to succeed, or auth fails
10:46 < major> the default seems to be: AuthenticationMethods publickey 
               password keyboard-interactive
10:47 < major> changing it to: publickey,password 
               publickey,keyboard-interactive means that you MUST pass 
               publickey authorization before you get to the password portion, 
               and then you can pass password or keyboard-interactive to 
               complete either of the 2 set definitions
10:48 < major> the TrustedUserCAKeys line means that the Host will not trust 
               any old public_key, only signed keys are trusted
10:48 < major> so users can not bipass security
10:48 < major> authorized_keys is "not" security
10:48 < major> as I said before
10:49 < major> again, the vast majority of the work is just signing public keys 
               and copy them back
10:49 < major> and I used a user_CA which is different than the host_CA
10:50 < major> one of the nice things about the system is that the signed cert 
               decides what usernames are allowed for attempted logins
10:51 < major> ssh-keygen -s users_CA -I user_amatus -n amatus,david,dbarksdale 
               -V +36w amatus_id.pub
10:51 < major> so there you have a signed cert that expires in 32 weeks, and 
               which can only be used to login as one of those 3 names
10:51 < major> the signing authority decides who the cert can be used to login 
               as
10:51 < major> none of this shit with "well, I have the password to someone 
               elses account"
10:53 < major> and again, it doesn't even matter if their pubkey is signed or 
               not
10:53 < major> because I am still forcing passwords
10:54 < major> it makes you wonder though .. how the fuck did people ever think 
               that authorized_keys and shit like sshguard or fail2ban were 
               solutions
10:54 < major> authorized_keys is almost akin to letting go of ones security 
               decisions
10:55 < major> its saying: we are not going to challange a user w/ a password, 
               we are going to trust our users to secure their public keys, we 
               are going to trust the security on their systems
10:55 < major> I dunno, doesn't matter
10:55 < major> ignorance abounds
10:55  * major shrugs.
11:03 < major> once this approach is all setup you just require that a user 
               send you a copy of their public cert and you sign it with the 
               list of names they are allowed to login as, and you send it back
11:03 < major> curiously, that keeps them from using a single id_rsa.pub for 
               all of their misc shit as well
11:06 < major> also, considering how much memory fail2ban can consume .. this 
               approach is a lot less memory intensive
11:07 < major> and protects against distributed attempts to attack the 
               passwords :P
11:07 < major> need I say moer?
11:07 < major> more* even
11:08 < major> and, the ability to expire a signed cert is just sort of a bonus